2024 Khan Text2CAD Sequential Parametric Models

Prototyping complex computer-aided design (CAD) models in modern software can be very time-consuming due to the lack of intelligent systems that can quickly generate simpler intermediate parts. In this study, we propose Text2CAD, the first AI framework for generating text-to-parametric CAD models designed for users of all skill levels. We introduce a data annotation pipeline to create text prompts based on natural language instructions for the DeepCAD dataset, which comprises approximately 170K models and 660K text annotations, from abstract CAD descriptions to detailed specifications. Within the Text2CAD framework, we present an end-to-end transformer-based auto-regressive network capable of generating parametric CAD models from input texts. Our model’s performance evaluation employs a mixture of metrics, including visual quality, parametric precision, and geometrical accuracy. The results indicate that our proposed framework demonstrates significant potential in AI-aided design applications, making the CAD modeling process more efficient and user-friendly. Project page is available at https://sadilkhan.github.io/text2cad-project.